Saving Our Bacon: Applications of Deep Learning for Precision Pig Farming

Abstract

The research presented in this thesis focussed on how deep learning can be applied to the field of agriculture to enable precision livestock farming for pigs. This refers to the use of technology to automatically monitor, predict, and manage livestock. Increased consumer awareness of the welfare issues facing animals in the farming industry, combined with growing demand for high-quality produce, has resulted in a need for providing farmers with tools to improve and simplify animal care. The concept of precision livestock farming tackles these requirements, as it makes it possible to treat animals as individuals, rather than as batches. This translates to tailored care for each animal and the potential for higher-quality produce. As deep learning has shown rapidly increasing potential in recent years, this research explored and evaluated various architectures for applications in two distinct areas within pig farming. We began by demonstrating how deep learning methods can be used to monitor and model the environmental conditions in which pigs are living in order to forecast oncoming respiratory disease. Implementing this approach can mean earlier intervention than if simplify looking for clinical symptoms. However, as not all diseases are caused by environmental conditions, we also implemented and evaluated a full workflow for the localisation and tracking of individual pigs. This made it possible to extract behavioural metrics to better understand the wellbeing of each pig. Overall, this research shows that deep learning can be used to advance the agriculture industry towards better levels of care, which is valuable for all stakeholders.